Summary

There aren't a lot of easy answers for what filesystem to use or how to arrange your disks. It's better to err on the side of paranoia, not performance, if you expect to keep your database intact under the sort of odd things that will happen to it in production. And unfortunately, maximum reliability is usually slower too. The most valuable thing you can do is be systematic in how you test. Instrument what you're doing, benchmark whenever possible, and always try to collect reproducible examples of realistic workloads whenever they appear.

• Filesystem crash protection is mostly commonly done by journaling writes, which adds overhead but makes recovery time after unclean shutdown predictable.
• On Linux, the ext3 filesystem allows a wide variety of tuning possibilities for its journal. While not the best performer, its occasional problem spots are at least well understood. Both XFS and ext4 are at early production quality right now, and expected to become increasingly...